yosarianilives

Benching with steponz

Need mem tab and can't use win 10 on this cpu for this bench

It'll need to get officially released first

Sounds like he's gonna be dropping some x265 gold soon then ?

Looks like I'm not the only one who thinks we should drop all legacy, we'll see how these percentages change as we get more votes.

Wow! Absolutely mental. I wonder how long before we see a sub 4 minute 32m score.

Are you sure it's the chip and not the board? I had a sabertooth devolop a -80c cbb but the same cpus fullpotted fine on other boards.

It physically a different socket lga 2011 vs lga 2011-1

Then we need someone to beat it

Current version uses whatever opencl driver uses and the newest intel opencl driver uses avx.

It's not just AVX, luumi had a 3770k die in the exact same way which does not have avx2. AVX units on ivy are very small compared to avx2 units on later arches.

I think you may be on to something here. This is probably gonna need to be tested more thoroughly tbh if we want to get to the bottom of why it's killing cpus. Anyone want to send me a tray of 9900k's?

I think people's complaint isn't that it's a chip killer so much that you can't tell what voltage it kills the chip because it's seemingly random and you won't know until the next day when the cpu no longer works. This is especially true when it's likely not even the AVX that's causing chip death. Ivy doesn't have avx2 yet luumi had his 3770k die in the exact same way as modern intel cpus are dying where it worked fine until the next day. I think that it's an issue with something else in how intel handles the instructions used by this encoder as anecdotally AMD cpus are not affected.

It may not be as simple as this as per my comment above.

I don't think this is AVX now, ivy bridge AVX is really light and doesn't really matter for this bench but Luumi apparently had his 3770k die in the exact same manner as everyone else's chips. Which points towards it being something else that's causing chip death in this bench, especially when AMD chips anecdotally seem to be immune. Also it sounds like the voltage to cause death varies by chip by a wide margin and is semi-random. I think this warrants further investigation as it may not be as simple as "run lower voltage than normal" and there could be something else going on here that's causing damage in this bench.

Bump, can I get an opinion on this @Leeghoofd

Why sub 7?

As far as I can tell what's causing the damage in X265 is the extreme power draw from AVX load on intel cpus. While it is weird that it doesn't kill it right away there's basically no other reason that this bench would kill a cpu. My guess is that the reason it's so much more harmful on intel cpus than amd cpus goes back to why there's varying AVX performance. The AVX units on an intel cpu are significantly bigger than on ryzen which is why they perform so much better. This also means that the avx power draw on ryzen is significantly below that on intel, especially when intel is using avx512. So I think it's not inconceivable to say that ryzen doesn't have the issue of dying the next day because the current draw isn't high enough to cause damage to the cpu. By the by, would you be able to run x265 on your 2990wx for Country Cup ?

I'd agree, if people want their hw to survive a heavy load they have to use a lower voltage. Already it's known that you don't run the same voltage for r15, x265, xtu etc as you would use for 32m or cpuz valid so perhaps we need to just use yet another offset from r15 for x265/xtu/whycruncher. The only thing is that it'll be really tempting to push the voltage just a bit harder for that max score since it won't die immediately like it would with something entirely stupid for other benches (eg 3v cpuz valid) so you'll be able to get a score.

That's the point of the revision, to simplify the code so that it actually works

Damnit, now I have to bench my chip. Good job man, absolutely mental clocks.

Then globals should be pulled from those benches.

If there's a wrapper then we'll need to drop all old scores as they won't be comparable to scores on the version with a wrapper.

Yeah changing in the middle of the week is not a bad idea, and my comments about sandbags were more towards those saying we shouldn't do background changes...

I'd put anything pre-dx11 as legacy, which for globals no card pre-dx11 would get points anyways so I don't see how it'd be a problem to drop globals for benches that even I can figure out how to glitch.